International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 3 Issue: 6

ISSN: 2321-8169
3799 - 3803

_______________________________________________________________________________________________

Implementation of Pre-processing and Efficient Blood Vessel Segmentation in

Retinopathy Fundus Image
Keerti V

Dr. Sarika Tale

M.Tech Student, Dept. of DECS

VTU-CPGS Bangalore Region
VIAT Muddenahalli,Chickaballapura,Karnataka,India
avc.1509@gmail.com

Associate Professor, Dept. of DECS

VTU-CPGS Bangalore Region
VIAT Muddenahalli,Chickaballapura,Karnataka,India
raga.sarika@gmail.com

Abstract- The human retina is a light receptive tissue and its enormously rich in blood vessels for its high physiological stress and dysfunction
of the retinal vasculature can effect from several diseases. Diabetic retinopathy is caused due to complications of diabetes, which can eventually
develop new blood vessels at the back of the retina and it can lead to blur vision or loss of vision. This work describes the problems of
retinopathy associated with diabetic patients and premature babies. We propose methods for the preprocessing and efficient segmentation
method to support measurement of the openness of the MTA, including image enhancement techniques like morphological operations, efficient
luminance component construction and bank of Gabor filters to segment retinal blood vessels. Finally an image cropping is used to separate
inferior and superior part of this segmented image for the effective and detailed analysis of the vascular structure in the fundus eye images.
Certain retinal disorders, if not detected in time, can cause serious problems like blur vision and blindness in patients. The implementation and
the performance of the various edge detection methods like Canny, Sobel and Gabor filters are based on visual perception. It has been concluded
that in case of natural images such as retinal fundus image a Gabor filter yielded better results in segmentation of blood vessels as compared to
edge detection methods of Canny and Sobel.
Keywords- Diabetic retinopathy, image preprocessing, Blood vessels, Gabor filter, Segmentation

__________________________________________________*****_________________________________________________
I.
INTRODUCTION
The transparency of the retina makes it in a direct window
for the human vasculature through an eye fundus
examination. Thus, some systemic diseases cause changes in
the retinal vessels which can be observed in retinal images.
Diabetic retinopathy (DR) is one of the most significant
complications of diabetes mellitus, which causes severe
damages in the retina, subsequently visual loss and
sometimes blindness if required medical treatment is not
taken on time. One of the difficulties in this illness is that
the patient with diabetes mellitus requires a continuous
screening for early detection. So far, various methods have
been planned by researchers to automate the recognition
process of retinal blood vessels in analyzing retinal images
such as color fundus images and fluorescein angiography
images of retinal fundus images. In this way, with the
increasing development of image processing techniques,
retina imaging constitutes a key for diagnosis support. The
problem of retinopathy is associated with diabetic patients
and premature babies. A signal processing and image
processing framework provides better diagnosis to the
Ophthalmologists for the analysis of the blood vessel
structure as accurately as possible. The vasculature of retinal
fundus is the only region of human body that can be
analyzed by using relatively simple and non-invasive
instrumentation . At the thresholding stage the accuracy of
the segmentation is dependent on the parameters used .

Alterations in the vascular structure of the retina can

indicate the presence of several types of pathology, such as
hypertension, arteriosclerosis, diabetes, myopia, and
retinopathy of prematurity (ROP). The vessels in the retina
are modified in terms of their width, shape, and tortuosity of
the retinal diseases. Variations in the architecture of the
major temporal arcade (MTA), in the form of a decrease in
the angle of insertion of the MTA as well as straightening of
the MTA, have been noted as manifestations of at least two
types of pathology: as a sequela of ROP and as an indicator
of the severity of myopia. In case of natural images such as
retinal fundus image a Gabor filter gives better results
compared to edge detection operators such as canny and
sobel but in case of medical images like CT scan , X-ray and
MRI, etc the Canny operator shows better results as it
provides fine details of an image. Canny edge detector
produces higher accuracy in detection of objects as
compared to Sobel [1,21].
II.

METHODOLGY

Below are the main steps involved in the implementation

and segmentation of the fundus image[21]
1) Preprocessing of images.
Normalizing each color component in the original
image.
Computing the luminance component.
3799

IJRITCC | June 2015, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 3 Issue: 6

ISSN: 2321-8169
3799 - 3803

_______________________________________________________________________________________________
Thresholding the luminance component to obtain the
effective area.
Expanding the luminance component beyond the
effective area to avoid the detection of its edges.

2) Obtaining the segmentation of the fundus image

Using 180 bank of Gabor filters over the range of
[90o, 90o] to represent the MTA.
Segmentation of the binary images.
Applying the morphological process to filter its
background.
Rotating each segmented image by 180o, if the
MTA opens to the left (i.e., the image is of the right
eye).
Obtaining the Gabor magnitude response image to
represent the STA and the ITA, respectively.
Cropping each segmented image horizontally.
Applying Canny, Sobel edge detectors to the
segmented image for comparative analysis.
Since the MATLAB is widely used interactive program
developed to perform numerical computation and data
visualization, our entire simulation program are written
using the programming language MATLAB.
A. Image preprocessing
In this method, the color scale RGB images are first
converted into grayscale for efficient computation. Since
the exudates are visible with more contrast in grayscale an
equalization procedure was performed on the images to
obtain a local contrast i,e approximately, equal at all image
intensities.
C (x, y) =(x, y) - median (x ,y)
(2.1)
Where C (x, y) is the estimated local contrast,
(x, y) is the image gray level at (x, y), median (x, y) is
the median gray level within the neighborhood
of (x
,y).This can be equated to a high pass spatial filter. The local
contrast provides a measure of the high-frequency image
noise[6].
B. Gabor filter
It acts as an excellent band pass filter. A complex Gabor
filter is defined as the product of Gaussian kernels times a
complex sinusoid.

2)
k,,f0 are filter parameters. The real and imaginary
components of Gabor filter are phase sensitive and allow
frequency tuning. This enables noise filtering and vessel

enhancement in single step. is spanned from 0o to 360o

over all four angles taken as pixel features. In [13], scale and
orientation of selective Gabor filter are used to detect and
classify the retina images into mild or severe case. The input
image is first filtered through Gabor filter banks, the banks
consist of several filters tuned to specific scales and
orientation and the operation is performed in frequency
domain, the output of which is the analyzed. Gabor filter
response is the convolution of the FT of the Gaussian
function and the Fourier transform of the harmonic function.
Gabor filter has a real and imaginary components
representing orthogonal directions. Thus image analysis
with Gabor filters is analogous to the human visual
perception system. Gabor filters are sinusoidally modulated
Gaussian functions. A Gabor filter is defined by the STD
valued of the Gaussian function in the X & Y directions ( x
& y ) and the frequency f0 of the modulating sinusoid as
[8].

(2.3)
As per design procedure, a variable named , representing
the average thickness of the vessels to be detected by Gabor
filters. We use = 16 pixels (0.32mm) f0 =1/ & x is in
relation to . A set of 180 Gabor filters spanning the range [
-90o , 90o] is prepared by rotating the
basic Gabor
function in equation . A magnitude response image is
created by using the maximum value of the response of
180gabor filters for each pixel. The morphological operation
of area open procedure also uses the concept of pixel
connectedness to detect segments of connected pixels
having less than 70 pixels and removes them. The main
principle of this stage is to eliminate background variations
in illumination from an image such that the forefront objects
are more easily analyzed. Therefore the binary image is
produced in which each pixel value is either 1 (blood vessel)
or 0 (background)[7].
III.

RESULTS AND DISCUSSION

The proposed blood vessel detection and segmentation

methodology is applied on images available in DRIVE
database and image collected from eye hospital. Hence, eye
care specialists can analyze retinopathy in larger populations
using this method. The results depicted in the below figures
are experimented with the images collected from the
standard Drive data base as well as the image collected
from the eye hospital.

3800
IJRITCC | June 2015, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 3 Issue: 6

ISSN: 2321-8169
3799 - 3803

_______________________________________________________________________________________________
the segmented result using Sobel method . From the above
result we observe that in DR eye fundus image, the blood
vessel structure has overgrowth and MTA has decreased.

A. Normal eye results

C. Severe Diabetic Retinopathy Results

(a)

(b)

using Canny method

using Sobel method

(a)

(b)
using Sobel method

(c )

(d )

(c)

(d)

Fig: 1 Results of normal eye

Fig 3: Severe Diabetic Retinopathy results

In above Fig 1: (a) indicates Gabor filter response of normal

eye (b) indicates the segmented image of Major Temporal
Arcade(MTA) Fig (c) depicts the segmented result using
Canny method and fig (d) shows the segmented result using
Sobel method .
From the above results, in normal eye fundus image the
blood vessel growth and MTA are normal.

In above Fig 3: (a) indicates Gabor filter response of severe

diabetic retinopathy(DR) (b) indicates the segmented image
of Major Temporal Arcade(MTA) Fig (c) depicts the
segmented result using Canny method and Fig (d) shows the
segmented result using Sobel method. In severe DR eye
fundus image, the blood vessel structure has severe
overgrowth and MTA has straightened.

B. Diabetic Retinopathy Results

(a)

The MTA diverges away from the Optic Nerve Head

towards the macula and then converges down into the
macular region. Furthermore, after the second or third
branching point, it becomes difficult to distinguish between
the original arcade and the new branch as they could be
similar in diameter and branch erratically. The second and
third branching points occur approximately over the
macula[22].
(b)

using Canny method

(c )
Fig 2: Diabetic Retinopathy results

using Sobel method

(d)

In above Fig 2: (a) indicates Gabor filter response of severe

diabetic retinopathy(DR) and (b) indicates the segmented
image of Major Temporal Arcade(MTA), Fig ( c) depicts
the segmented result using Canny method and fig (d) shows

A. Comparative analysis:
From the above results, we observe that Sobel operator has
poor edge localization in segmented image. It is sensitive to
noise and gives inaccurate results. Hence it is not an
efficient method in analysis of the segmented blood vessels.
Canny edge detector has adaptive thresholding and produces
higher accuracy in detection of objects as compared to Sobel
but Canny operator is time consuming and has complex
computation. To overcome the above problem a filter bank
consisting of 180 Gabor filters with cosine term having
various scales and rotations of kernel functions are used.
The Gabor Filter is very useful in image processing
applications using edge detection and provides optimal
localization in both the frequency and space domains. Gabor
filter has better feature detection and produces thick edges.
3801

IJRITCC | June 2015, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 3 Issue: 6

ISSN: 2321-8169
3799 - 3803

_______________________________________________________________________________________________
Therefore Gabor filter gives better results as compared to
Canny and Sobel in analyzing the segmented blood vessels
of retinopathy with improved signal to noise ratio.
IV.

CONCLUSION

Diabetic retinopathy is the leading cause of the blindness in

diabetic populations. The effective segmentation of the eye
funds image helps to detect and diagnose to help prognosiss
treatment procedural plans if required for the
ophthalmologists. In this work we have presented diabetic
image segmentation technique using bank of Gabor filters
with effective image enhancement technique like
normalized gray scale conversion, masking and optic nerve
head elimination. This framework also provides closer
analysis of the superior and inferior arcade portions by
cropping superior and inferior portions from the segmented
image. The experimented results shows that the
effectiveness of the image segmentation algorithm using
bank of Gabor filters to keep only the blood vessels
structures and avoids other artifacts. Generally Gabor filter
gives better results than Canny and Sobel especially for
natural images, where the objects are considered to be a big
challenge to segment. Thus Gabor filter provides better
result as the selected parameters produce the large
magnitude and brighter intensity of the edges.
V.

FUTURE SCOPE

Automatic retinopathy detection is a primary challenge in

the todays world because of the feature extraction of the
retinopathy images with and without diabetic retinopathy is
very complicated using image processing techniques.
Temporal screening of the fundus image is not more often
recommended by the physicians. Therefore temporal feature
extraction, regular temporal screening of fundus image and
machine learning based automatic retinopathy detection is
recommended as the future scope from this work.

[3]
[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

[13]

ACKNOWLEDGMENT
Thanks to my Guide, family member and my friend
Satyanarayana V who always support, help and guide during
my dissertation work. Finally thanks to Dr.Bhargavi,
consulting physician, Minto Eye Hospital Bangalore for
providing necessary information about retinopathy.

[14]
[15]

REFERENCES
[1] International Committee for the Classification of
Retinopathy
of
Pre-maturity.The
international
classification of retinopathy of prematurity revisited.
Archives of ophthalmology, 123:991999, 2005.
[2] Wilson CM, Cocker KD, Moseley MJ, Paterson C, Clay
ST, Schulenburg WE, Mills MD, Ells AL, Parker KH,

[16]

Quinn GE, Fielder AR, and Ng J. Computerized analysis

of retinal vessel width and tortuosity in premature
infants. Investigative Ophthalmology and Visual
Science, 49(1):3577 3585, 2008.
Jelinek HF and Cree MJ. Automated Image Detection of
Retinal Pathology. CRC Press, Boca Raton, FL, 2010.
Patton N, Aslam TM, MacGillivray T, Deary IJ, Dhillon
B, Eikelboom RH, Yogesan K, and Constable IJ. Retinal
image analysis: Concepts, applications and potential.
Progress in Retinal and Eye Research, 25(1):99127,
2006.
Fledelius HC and Goldschmidt E. Optic disc appearance
and retinal temporal vessel arcade geometry in high
myopia, as based on follow-up data over 38 years. Acta
Ophthalmologica, 88(5):514520, March 2010.
Wilson C, Theodorou M, Cocker KD, and Fielder AR.
The temporal retinal vessel angle and infants born
preterm. British Journal of Ophthalmology, 90:702704,
2006.
Reese AB, King MJ, and Owens WC. A classification of
retrolental
fibroplasia.
American
Journal
of
Ophthalmology, 36:13331335, 1953.
Cryotherapy for Retinopathy of Prematurity Cooperative
Group. Multi-center trial of cryotherapy for retinopathy
of prematurity: Ophthalmolog-ical outcomes at 10 years.
Archives of Ophthalmology, 119:11101118, 2001.
Foracchia M, Grisan E, and Ruggeri A. Detection of
optic disc in retinal images by means of a geometrical
model of vessel structure. IEEE Transactions on Medical
Imaging, 23(10):11891195, 2004.
Tobin KW, Chaum E, Govindasamy VP, and Karnowski
TP. Detection of anatomic structures in human retinal
imagery. IEEE Transactions on Medical Imaging,
26(12):17291739, December 2007.
Li H and Chutatape O. Automated feature extraction in
color retinal images by a model based approach. IEEE
Transactions on Biomedical Engineering, 51(2):246254,
2004.
Fleming AD, Goatman KA, Philip S, Olson JA, and
Sharp PF. Automatic detection of retinal anatomy to
assist diabetic retinopathy screening. Physics in
Medicine and Biology, 52:331345, 2007.
D. Vallabha, R. Dorairaj, K.R. Namuduri and
H.Thompson, Automated Detection and Classification
of
Vascular
Abnormalities
in
Diabetic
Retinopathy,Conference Record of the 38th Asilomar
Conference on Signals, Systems and Computers, Pacific
Grove, CA,USA,vol.2,pp.1625-1629,Nov.2004.
R. O. Duda, P. E. Hart, and D. G. Stork, Pattern
Classification, 2nd ed. New York: Wiley, 2001
J. Hayashi, T. Kunieda, J. Cole, R. Soga, Y. Hatanaka,
M.Lu, T. Hara and H. Fujita, A Development of
Computer Aided Diagnosis System Using Fundus
Images, Proceeding of the 7th International Conference
on Virtual Systems and Multi Media, Berkeley,
California,USA,pp.429-438,Oct.2001.
Ying H and Liu JC. Automated localization of maculafovea area on retina images using blood vessel network
topology. In Acoustics Speech and Signal Processing
3802

IJRITCC | June 2015, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 3 Issue: 6

ISSN: 2321-8169
3799 - 3803

_______________________________________________________________________________________________
[17]

[18]

[19]

[20]
[21]

[22]

[23]

[24]

[25]

(ICASSP), 2010 IEEE International Conference on,

pages 650653, March 2010 .
Balmer A, Zografos L, Munier F; Diagnosis and current
management of retinoblastoma. Oncogene. 2006 Aug
28;25(38):5341-9.
Oloumi F, Rangayyan RM, and Ells AL. Parametric
representation of the retinal temporal arcade. In The 10th
IEEE International Conference on Information
Technology and Applications in Biomedicine, Corfu,
Greece, November 2010.
DRIVE: Digital Retinal Images for Vessel Extraction.
http:// www. isi.uu.nl/ Research/ Databases/ DRIVE/,
accessed on March 24, 2008.
Image Processing and Analysis in Java, http:// rsbweb.
nih.gov/ ij/, accessed on September 3, 2008.
Rangayyan RM, Ayres FJ, Oloumi F, Oloumi F, and
Eshghzadeh-Zanjani P. Detection of blood vessels in the
retina with multiscale Gabor filters. Journal of Electronic
Imaging, 17:023018:17, April-June 2008.
Rangayyan RM, Zhu X, Ayres FJ, and Ells AL.
Detection of the optic nerve head in fundus images of the
retina with Gabor filters and phase portrait analysis.
Journal of Digital Imaging, 23(4):438453, August 2010.
Sample
Image,[Online],Avaliable,
http://www.dambrosio-eye-care-boston.com/diabeticretinopathy.html, Aug. 2007.
M. Niemeijer, M. D. Abramoff, and B. v. Ginneken,
Segmentation of the optic disc, macula and vascular
arch in fundus photographs, IEEE Trans. Med. Imag.,
vol. 26, no. 1, pp. 116127, Jan. 2007.
F. Zana and J. Klein, Segmentation of vessel-like
patterns using mathematical morphology and curvature
evaluation, IEEE Trans. Image Process., vol. 10, no. 7,
pp. 10101019, Jul. 2001.
L. Gagnon, M. Lalonde, M. Beaulieu, and M.-C.
Boucher, Procedure to detect anatomical structures in
optical fundus images in Proc. SPIE Med. Imaging:
Image Processing, vol. 4322, 2001,pp. 12181225.

3803
IJRITCC | June 2015, Available @ http://www.ijritcc.org

_______________________________________________________________________________________